vc4: Cache the texture p1 for the sampler.

Cuts another 12% of vc4_uniforms.o, in exchange for computing it at
CSO creation time.

3 files changed, 69 insertions, 49 deletions

diff --git a/src/gallium/drivers/vc4/vc4_context.h b/src/gallium/drivers/vc4/vc4_context.h
index be7df1bd34..7faf522363 100644
--- a/src/gallium/drivers/vc4/vc4_context.h
+++ b/src/gallium/drivers/vc4/vc4_context.h
@@ -75,6 +75,11 @@ struct vc4_sampler_view {
         uint32_t texture_p1;
 };
 
+struct vc4_sampler_state {
+        struct pipe_sampler_state base;
+        uint32_t texture_p1;
+};
+
 struct vc4_texture_stateobj {
         struct pipe_sampler_view *textures[PIPE_MAX_SAMPLERS];
         unsigned num_textures;
@@ -338,6 +343,12 @@ vc4_sampler_view(struct pipe_sampler_view *psview)
         return (struct vc4_sampler_view *)psview;
 }
 
+static inline struct vc4_sampler_state *
+vc4_sampler_state(struct pipe_sampler_state *psampler)
+{
+        return (struct vc4_sampler_state *)psampler;
+}
+
 struct pipe_context *vc4_context_create(struct pipe_screen *pscreen,
                                         void *priv);
 void vc4_draw_init(struct pipe_context *pctx);
diff --git a/src/gallium/drivers/vc4/vc4_state.c b/src/gallium/drivers/vc4/vc4_state.c
index 46852b6628..8a759c2ca4 100644
--- a/src/gallium/drivers/vc4/vc4_state.c
+++ b/src/gallium/drivers/vc4/vc4_state.c
@@ -461,11 +461,64 @@ vc4_get_stage_tex(struct vc4_context *vc4, unsigned shader)
         }
 }
 
+static uint32_t translate_wrap(uint32_t p_wrap, bool using_nearest)
+{
+        switch (p_wrap) {
+        case PIPE_TEX_WRAP_REPEAT:
+                return 0;
+        case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+                return 1;
+        case PIPE_TEX_WRAP_MIRROR_REPEAT:
+                return 2;
+        case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+                return 3;
+        case PIPE_TEX_WRAP_CLAMP:
+                return (using_nearest ? 1 : 3);
+        default:
+                fprintf(stderr, "Unknown wrap mode %d\n", p_wrap);
+                assert(!"not reached");
+                return 0;
+        }
+}
+
 static void *
 vc4_create_sampler_state(struct pipe_context *pctx,
                          const struct pipe_sampler_state *cso)
 {
-        return vc4_generic_cso_state_create(cso, sizeof(*cso));
+        static const uint8_t minfilter_map[6] = {
+                VC4_TEX_P1_MINFILT_NEAR_MIP_NEAR,
+                VC4_TEX_P1_MINFILT_LIN_MIP_NEAR,
+                VC4_TEX_P1_MINFILT_NEAR_MIP_LIN,
+                VC4_TEX_P1_MINFILT_LIN_MIP_LIN,
+                VC4_TEX_P1_MINFILT_NEAREST,
+                VC4_TEX_P1_MINFILT_LINEAR,
+        };
+        static const uint32_t magfilter_map[] = {
+                [PIPE_TEX_FILTER_NEAREST] = VC4_TEX_P1_MAGFILT_NEAREST,
+                [PIPE_TEX_FILTER_LINEAR] = VC4_TEX_P1_MAGFILT_LINEAR,
+        };
+        bool either_nearest =
+                (cso->mag_img_filter == PIPE_TEX_MIPFILTER_NEAREST ||
+                 cso->min_img_filter == PIPE_TEX_MIPFILTER_NEAREST);
+        struct vc4_sampler_state *so = CALLOC_STRUCT(vc4_sampler_state);
+
+        if (!so)
+                return NULL;
+
+        memcpy(so, cso, sizeof(*cso));
+
+        so->texture_p1 =
+                (VC4_SET_FIELD(magfilter_map[cso->mag_img_filter],
+                               VC4_TEX_P1_MAGFILT) |
+                 VC4_SET_FIELD(minfilter_map[cso->min_mip_filter * 2 +
+                                             cso->min_img_filter],
+                               VC4_TEX_P1_MINFILT) |
+                 VC4_SET_FIELD(translate_wrap(cso->wrap_s, either_nearest),
+                               VC4_TEX_P1_WRAP_S) |
+                 VC4_SET_FIELD(translate_wrap(cso->wrap_t, either_nearest),
+                               VC4_TEX_P1_WRAP_T));
+
+        return so;
 }
 
 static void
diff --git a/src/gallium/drivers/vc4/vc4_uniforms.c b/src/gallium/drivers/vc4/vc4_uniforms.c
index e8e9d6b011..5613d6b28c 100644
--- a/src/gallium/drivers/vc4/vc4_uniforms.c
+++ b/src/gallium/drivers/vc4/vc4_uniforms.c
@@ -27,26 +27,6 @@
 #include "vc4_context.h"
 #include "vc4_qir.h"
 
-static uint32_t translate_wrap(uint32_t p_wrap, bool using_nearest)
-{
-        switch (p_wrap) {
-        case PIPE_TEX_WRAP_REPEAT:
-                return 0;
-        case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-                return 1;
-        case PIPE_TEX_WRAP_MIRROR_REPEAT:
-                return 2;
-        case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-                return 3;
-        case PIPE_TEX_WRAP_CLAMP:
-                return (using_nearest ? 1 : 3);
-        default:
-                fprintf(stderr, "Unknown wrap mode %d\n", p_wrap);
-                assert(!"not reached");
-                return 0;
-        }
-}
-
 static void
 write_texture_p0(struct vc4_context *vc4,
                  struct vc4_cl_out **uniforms,
@@ -68,34 +48,10 @@ write_texture_p1(struct vc4_context *vc4,
 {
         struct vc4_sampler_view *sview =
                 vc4_sampler_view(texstate->textures[unit]);
-        struct pipe_sampler_state *sampler = texstate->samplers[unit];
-        static const uint8_t minfilter_map[6] = {
-                VC4_TEX_P1_MINFILT_NEAR_MIP_NEAR,
-                VC4_TEX_P1_MINFILT_LIN_MIP_NEAR,
-                VC4_TEX_P1_MINFILT_NEAR_MIP_LIN,
-                VC4_TEX_P1_MINFILT_LIN_MIP_LIN,
-                VC4_TEX_P1_MINFILT_NEAREST,
-                VC4_TEX_P1_MINFILT_LINEAR,
-        };
-        static const uint32_t magfilter_map[] = {
-                [PIPE_TEX_FILTER_NEAREST] = VC4_TEX_P1_MAGFILT_NEAREST,
-                [PIPE_TEX_FILTER_LINEAR] = VC4_TEX_P1_MAGFILT_LINEAR,
-        };
-
-        bool either_nearest =
-                (sampler->mag_img_filter == PIPE_TEX_MIPFILTER_NEAREST ||
-                 sampler->min_img_filter == PIPE_TEX_MIPFILTER_NEAREST);
-
-        cl_aligned_u32(uniforms, sview->texture_p1 |
-               VC4_SET_FIELD(magfilter_map[sampler->mag_img_filter],
-                             VC4_TEX_P1_MAGFILT) |
-               VC4_SET_FIELD(minfilter_map[sampler->min_mip_filter * 2 +
-                                           sampler->min_img_filter],
-                             VC4_TEX_P1_MINFILT) |
-               VC4_SET_FIELD(translate_wrap(sampler->wrap_s, either_nearest),
-                             VC4_TEX_P1_WRAP_S) |
-               VC4_SET_FIELD(translate_wrap(sampler->wrap_t, either_nearest),
-                             VC4_TEX_P1_WRAP_T));
+        struct vc4_sampler_state *sampler =
+                vc4_sampler_state(texstate->samplers[unit]);
+
+        cl_aligned_u32(uniforms, sview->texture_p1 | sampler->texture_p1);
 }
 
 static void